It has been proposed that biodiversity can be important for ecosystem functioning and act as an insurance against perturbations and environmental fluctuations. To date, theoretical work supports this idea but direct experimental evidence is still to some extent ambiguous and debated. The main reason for this debate – and the lack of strong empirical support – is due to unavoidable experimentally and statistically inherent variance reduction effects. Here we present the results of an experimental study that circumvents earlier hidden treatments. By random draw without replacement, we collected 180 full-sibling batches of an amphipod from a large pool of possible parents. Assembled amphipod populations with diversity levels ranging from one to ten were exposed to either a single perturbation (nutrient enrichment) or two combined perturbations (nutrient enrichment and desiccation). The results show that the variance in the number of surviving individuals decreased with increasing diversity in the combined perturbations treatment. Predictability in population survival thus seemed to be higher in more diverse assemblages. Our results, together with a simple model suggest that variance-decreasing effects can be due to actual real world statistical sampling effects of increasing diversity.